Tag
The paper diagnoses failures in test-time reinforcement learning for medical QA due to answer-space structure and introduces PROSE, which rewards reasoning quality to improve model performance without labeled data.
SERPO introduces a self-evolving rubric policy optimization framework for test-time reinforcement learning in open-ended generation, replacing answer voting with a closed loop that co-evolves response evidence, query-specific rubrics, and policy parameters, achieving significant improvements on health and research benchmarks.
Introduces Test-Time Reinforcement Learning (TTRL), a method that uses majority voting on unlabeled data to create pseudo-labels for RL training, enabling self-improvement of LLMs without ground-truth answers. Achieves significant gains (e.g., +159-211% on AIME 2024 for Qwen-2.5-Math-7B).